Autonomous Learning of Bipedal Walking Stabilization

人形机器人，具有类似人的身体计划的机器人，作为研究平台越来越受欢迎，因为它们有潜力成为通用机器人，可以在适合人类需求的环境中执行各种各样的任务。然而，复制人类自然步态的效率、健壮性和优雅性是一个巨大的挑战。两足行走者天生就不稳定，难以控制。在优先计划自主学习的第一个资助期，我们开发了一种新颖的闭环控制方法，用于存在大干扰的双足行走。我们的捕获步进控制器能够吸收在一步中任何方向和任何时间发生的干扰，并在一到两步内使机器人恢复到所需的行走速度。为了使第一个捕获步骤更加有效，我们开发了一种方法，以在线方式学习描述与点质量模型偏差的非参数模型-从采取捕获步骤后的少量剩余能量观测中。对于SPP的第二个资助期，我们建议研究超越捕获步骤的更先进的推力恢复策略，即利用躯干的旋转惯性来补充捕获步骤和平衡恢复扭矩。此外，我们建议开发在线调整步骤的方法，以适应视觉感知的足部放置限制。同样，我们的平衡控制器的参数将以非参数的方式在线学习，更大的状态-作用空间将被自主探索。这些发展将扩展我们的平衡控制方法的适用性，以限制足部放置的受限环境，并保持平衡的情况远远超出了现有的推恢复方法的限制。

Humanoid robots, robots with a human-like body plan, enjoy increasing popularity as research platform, because they have the potential to be universal robots that can perform a large variety of tasks in environments that are designed to suit human needs.It is, however, a big challenge to replicate the efficiency, robustness, and grace of the natural human gait. Bipedal walkers are inherently unstable and difficult to control. In the first funding period of the Priority Programme Autonomous Learning, we developed a novel, closed-loop control approach for bipedal walking in the presence of large disturbances. Our capture step controller is able to absorb disturbances that can occur from any direction and at any time during a step and returns the robot to the desired walking velocity within one or two steps. To make the first capture step even more effective, we developed a method to learn in an online fashion a non-parametric model describing deviations from the point-mass model - from a small number of observations of residual energy after taking capture steps. For the second funding period of the SPP, we propose to investigate more advanced push recovery strategies that go beyond capture steps, i.e., using the rotational inertia of the torso to complement capture steps with balance-restoring torque. Furthermore, we propose the development of methods for online adaptation of steps to visually perceived constraints on foot placement. Again, parameters of our balance controllers will be learned online in a non-parametric way and the larger state-action space will be explored autonomously. These developments will extend the applicability of our balance control methods to restricted environments with constraints on foot placement, and also maintain balance in situations well beyond the limits of existing push recovery approaches.